Nikola Novaković odbranio je rad na MCF programu na temu „Credit scoring and default prediction“

Student Nikola Novaković odbranio je master rad na temu Credit scoring and default prediction, pred komisijom koju su činili mentor prof. dr Branko Urošević, kao i članovi prof. dr Milan Nedeljković i prof. dr Vladimir Vasić. Nikola je odbranom master rada završio studije na programu MCF (Master in Computational Finance).

U uvodu svog rada Nikola je istakao:

„An in-depth analysis of the development of credit scoring models, as well as their current and past states, regulations, success indicators, validation methods and perspectives for further study is carried out in this master’s thesis. Chapter one of the research delves into the historical evolution of credit scoring. It starts with the rudimentary judgmental methods before leading to the contemporary statistical models. The importance of adaptive risk management in a dynamic financial space is highlighted.

In Chapter 2, we discuss traditional credit scoring models, emphasizing the continued weight in benchmarks such as the FICO score. On the other hand, Chapter 3 examines the development of contemporary credit scoring techniques which incorporate machine learning and the utilization of massive datasets to achieve detailed risk rating.

Chapter 4 focuses on the regulatory considerations surrounding credit scoring and looks at the changing regulation surrounding the area. The way forward in this regard entails striking an appropriate balance between encouraging innovation and protecting consumer interests.

Chapter 5 describes how accurate, precise, and recall work within credit scoring model. Furthermore, how ROC is vital in a credit-scoring model is also explained. Such measures work out as the compasses directing the construction and evaluating of models in a widening monetary environment.

Chapter 6 further investigates validation methods and ascertains that verifying reliability and strength of credit scoring models requires validation techniques. Hence, it is imperative to have a rigorous validation process, ranging from traditional holdout samples to advanced cross-validation methods, to avoid overfitting and ensure that models are generalizable.

Chapter 7 looks forward and considers possible directions for credit scoring in the future. The inclusion of explainable AI and alternative data suggest that credit scoring will in the near future give explainable and fair explanations on why it predicts risk precisely.

Therefore, through the combination of historical knowledge and advanced methods, this thesis proves the necessity of a thorough and well-grounded credit scoring. The interplay of these elements make a credit scoring environment robust, reliable, honest and fair. The concluding thoughts provide guidance to practitioners, policy makers, and scholars on how to make credit scoring more human.“


„In the end, it is evident that the history behind credit scoring is indeed a colorful mix of the threads of history, tradition, innovation, and regulation. We have shown the unyielding will to achieve better risk pricing from our earliest valuation methodologies to today’s advanced statistical models in an ever-changing and ageless financial environment.

Going back to basics, Chapter 2 took us back to the traditional credit scoring models such as the trustworthy FICO score. These models have formed the main basis for most of the credit decisions for decades, and offer reliable basis.

However, as we have come to learn from Chapter 3, new credit scoring techniques are rewriting the story. Lenders can now utilize machine learning and big data to get a better understanding of data that comes from a wide cross section of users.

In essence, it was in Chapter 4 that we managed to sail through the murky waters of credit rating regulation. It is increasingly vital to strike a healthy equilibrium between encouraging innovation and safeguarding consumers these days. Since technology is changing, the credit scoring rules should also change to be fair, transparent, and honest when treating customer data.

Chapter 5 elaborated the vital numbers in the credit rating model. These will be our guides as we navigate the world of credit scoring from accuracy to ROC curves. These act as checkboxes to make sure we are on track and to keep bias and misinformation from happening.

Chapter 6 was on validation techniques which are crucial in the world. It is just like conducting a strain test of a bridge and then allowing people to pass over it. It is highly imperative that traditional holdout samples or even modern cross-validation techniques are used to rigorously test these models.

The chapter 7 offered a glimpse on the future directions that may be adopted for the credit scoring system. Risk forecasting is not only about right prediction. It is essential for the projections to be sensible, justifiable, and accurate. The emergence of explainable AI and using of alternative data sources suggest that soon credit scores will be more than just numerical values because they will tell comprehensible stories.

Indeed, this work is not simply a bunch of chapters. For any novice navigating the maze world of credit scoring, this is a guide. It integrates what people learned in past, understandings of today, and outlook for tomorrow. Credit scoring using machine intelligence in the future will need a blend of expertise of man and accountability that will be transparent. The ideas discussed within chapters summarize and add some inputs into the debate concerning the direction of credit scoring.“ – zaključio je Nikola.

Fotografije sa odbrane dostupne su u galeriji.